An ERG gene ‘break-apart’ fluorescence in situ hybridization (FISH) assay has been used to screen whole-mount prostatectomy specimens for rearrangements at the ERG locus. In cancers containing ERG alterations the observed pattern of changes was often complex. Different categories of ERG gene alteration were found either together in a single cancerous region or within separate foci of cancer in the same prostate slice. In some cases the juxtaposition of particular patterns of ERG alterations suggested possible mechanisms of tumour progression. Prostates harbouring ERG alterations commonly also contained cancer that lacked rearrangements of the ERG gene. A single trans-urethral resection of the prostate specimen examined harboured both ERG and ETV1 gene rearrangements demonstrating that the observed complexity may, at least in part, be explained by multiple ETS gene alterations arising independently in a single prostate. In a search for possible precursor lesions clonal ERG rearrangements were found both in high grade prostatic intraepithelial neoplasia (PIN) and in atypical in situ epithelial lesions consistent with the diagnosis of low grade PIN. Our observations support the view that ERG gene alterations represent an initiating event that promotes clonal expansion initially to form regions of epithelial atypia. The complex patterns of ERG alteration found in prostatectomy specimens have important implications for the design of experiments investigating the clinical significance and mechanism of development of individual prostate cancers.
Fusion of the TMPRSS2 gene to the ETS family transcription factor gene ERG has been reported as a common event in human prostate cancer (Tomlins et al., 2005, 2006; Clark et al., 2006; Hermans et al., 2006; Iljin et al., 2006; Perner et al., 2006; Wang et al., 2006; Yoshimoto et al., 2006). Less frequently TMPRSS2 also becomes fused to ETV1 and ETV4 (Tomlins et al., 2005, 2006; Hermans et al., 2006). Formation of a TMPRSS2–ERG fusion gene leads to disruption of the ERG gene locus which can be detected by fluorescence in situ hybridization (FISH) assays that measure the separation of 5′- and 3′-ERG sequences (Tomlins et al., 2005; Perner et al., 2006). FISH assays detect ERG translocations in up to 50% of prostate cancer cases, in approximate agreement with values of 55 and 59% (Tomlins et al., 2005; Wang et al., 2006) obtained using reverse transcription (RT–PCR-based detection of TMPRSS2–ERG fusions. The TMPRSS2 and ERG genes are positioned in the same orientation on chromosome 21 and 2.85 Mb apart, and it has been shown by several groups (Hermans et al., 2006; Iljin et al., 2006; Perner et al., 2006) that TMPRSS2–ERG fusions are frequently accompanied by loss of the entire intervening chromosome 21 sequence. Such deletions are found in about 60% of cancers containing an ERG rearrangement (Perner et al., 2006).
Several studies have demonstrated a relationship between ERG gene status and clinicopathological indicators (Perner et al., 2006; Wang et al., 2006; Demichelis et al., 2007; Nam et al., 2007). For example, Perner et al. (2006) found that the presence of an ERG rearrangement accompanied by 5′-ERG deletion had a significant correlation with higher tumour stage and with the presence of metastatic disease involving pelvic lymph nodes. In a watchful waiting cohort of 111 patients, Demichelis et al. (2007) reported both a significant association between the presence of a TMPRSS2–ERG fusion and prostate cancer-specific death and a link between the presence of ERG alterations and higher Gleason score. In a series of 26 Gleason seven patients who underwent prostatectomy, Nam et al. (2007) found that the presence of a TMPRSS2–ERG fusion was associated with a greater probability of biochemical disease relapse (elevated prostate-specific antigen, PSA). We have recently determined that loss of 5′-ERG sequences coupled with duplication of translocated ERG predicts extremely poor clinical outcome (Attard et al., 2007) independently of Gleason score and PSA level at diagnosis.
The results of these studies imply that cancerous prostates can be divided into two groups: those that contain an ERG rearrangement and have poorer clinicopathological indictors and those that lack an ERG rearrangement and have a more favourable outcome. Unsupervised clustering analyses of prostate cancer expression microarray data sets do not support this view because such analyses have currently failed to identify consistently distinct cancer categories (Singh et al., 2002; Lapointe et al., 2004; Yu et al., 2004; Stephenson et al., 2005) unlike, breast cancer, for example, where separate basal, luminal and ERBB2 subgroups are recognized (Sorlie et al., 2001, 2003). However, a limitation of prostate cancer expression microarray profiles is that they are based on sampling procedures that do not take into account either the well-documented occurrence of multi-focal disease (Villers et al., 1992; Aihara et al., 1994; Miller and Cygan, 1994; Djavan et al., 1999; Chen et al., 2000; Arora et al., 2004) or the existence of genetic heterogeneity found within individual foci of cancer (Konishi et al., 1995; Mirchandani et al., 1995; Qian et al., 1995; Jenkins et al., 1997; Bostwick et al., 1998; Cheng et al., 1998). ERG gene alterations have been found both in prostate cancer (Tomlins et al., 2005) and in high grade prostatic intraepithelial neoplasia (HG-PIN) (Cerveira et al., 2006; Perner et al., 2007). In a recent RT–PCR-based screen we detected TMPRSS2–ERG fusion transcripts in samples selected from apparently non-neoplastic regions of cancerous prostates, although we could not formally exclude the possibility that a fusion was present in small regions of cancer present in the selected samples (Clark et al., 2006). These observations prompted us to use a FISH ‘break-apart’ assay to carry out a screen for rearrangements of the ERG gene in a series of whole-mount prostate cancers sections, with the objective of assessing heterogeneity of ERG gene status within a single prostate and of gaining further insights into the mechanism of development of human prostate cancer.
Detection of ERG rearrangements in TMAs by FISH
We initially constructed tissue microarrays (TMAs) with cancer cores selected from the major areas of cancer identified in each of a series of formalin-fixed prostatectomy specimens; 82 cancer cores were analysed from 31 patients. To assess the presence of the TMPRSS2–ERG fusions in this cancer set we used an ERG gene break-apart assay similar to that described by Tomlins et al. (2005) using probes 1 and 2 (Figure 1). A total of 58% (18 of 31) were found to harbour an ERG rearrangement. Based on the observed FISH patterns, cancers were stratified according to whether they were (1) class N (for normal) with un-rearranged ERG loci characterized by twinned red (5′-ERG) and green (3′-ERG) FISH signals (13 cancers, 42%; Figure 1a), (2) class ‘Esplit’ which had a rearranged ERG locus manifested as separation of the red and green signals (6 cancers, 19%; Figure 1b) or (3) class ‘Edel’ (for ERG 5′ deletion), an ERG rearrangement with loss of green signal corresponding to 5′-ERG sequences (12 cancers, 39%; Figures 1c and d). Within Edel the pattern could be further divided into cancers that contained one copy of the ‘red’ 3′-ERG signal (‘1Edel’; Figure 1c) and those that exhibit more than one copy of the 3′-ERG FISH signal (‘2+Edel’; Figure 1d). Re-hybridization of the slices exhibiting ERG rearrangement to a 5′-TMPRSS2 probe (described in Attard et al., 2007) confirmed 3′-ERG sequences remained juxtapositioned to 5′-TMPRSS2 sequences despite loss (Edel) or relocation (Esplit) of the intervening sequences (represented by probe 2) between TMPRSS2 and ERG. RT–PCR studies performed as previously described (Clark et al., 2006) confirmed that class Esplit and class Edel ERG samples detected by FISH did indeed express 5′-TMPRSS2–ERG-3′ fusion transcripts (results not shown).
We have recently reported a link between clinical outcome and certain FISH signal copy number variations (Attard et al., 2007). Consequently we have recorded the precise number of un-rearranged ERG (n=1–4), 3′-ERG (red, n=0–2) and 5′-ERG (green, n=0–2) signals observed within each category. To record this variation the inter-phase nuclei were scored according to the number of each of the three FISH signals (A, B and C) where, ‘A’ is the number of twinned red and green signals corresponding to the wild-type ERG locus; ‘B’ the number of lone red signals corresponding to translocated 3′-ERG and ‘C’ the number of lone green signals corresponding to translocated 5′-ERG. A cancer with an (A, B and C) score of (1, 2 and 0) is, for example, a class ‘Edel’ cancer that contains one un-rearranged ERG allele, two separate red 3′-ERG signals and no separate green 5′-ERG signal.
ERG rearrangements in whole-mount prostatectomy sections
For these analyses we randomly selected five prostatectomy specimens that had been designated ERG rearrangement positive in the initial TMA-based study (see above) and four cancers that had been designated ERG rearrangement negative. These cancers are listed in Table 1. The FISH ERG break-apart assay was used to systematically screen all cells within a single 4 μm section of each prostate for rearrangements of the ERG gene. Flanking adjacent sections were subject to (1) haematoxylin and eosin (H&E) staining, and (2) a combined immunohistochemical stain for p63 and α-methylacyl-CoA racemase (AMACR). p63 highlights normal basal cell nuclei, while relative over-expression of AMACR can aid a diagnosis of prostate neoplasia). For each prostate sample we mapped (1) the ERG gene status of each region of cancer and HG-PIN and (2) any other area that contained abnormalities at the ERG locus. Maps of the ERG alterations identified in these studies are shown at low resolution in Figure 4 and at high resolution in Figures 2 and 3 and in Supplementary Figures 1–9. Clinical information for the samples plus FISH ERG status of the whole-mount prostatectomy and corresponding original TMA cores is provided in Table 1. As expected, ERG alterations were entirely restricted to cells with an epithelial morphology, and were not detected in stromal cells.
ERG rearrangements in cancer
A number of observations emerged from these analyses. Most of the prostates examined (6 of 9) appeared to be multi-focal, harbouring distinct regions of cancer or ERG alteration in separate regions of the same prostate: only cancers 908, 8341 and 5170 were not multi-focal. Three of the five cancers originally designated ERG rearrangement positive in TMA studies also contained cancer that lacked an ERG alteration (2007, 908 and 543) thus overall seven of the nine prostates examined (all prostates except 1266, 8341; Figure 4) contained class N cancer.
We failed to find any evidence of ERG rearrangements in prostates 5170, 6972 and 15481 (Figure 4). Prostate 2243 contained three separate regions of cancer. Two of these harboured un-rearranged ERG loci but the third contained class Esplit ERG alterations. Interestingly one of the regions of class N cancer had an area of cancer with two copies of ERG juxtapositioned to areas of cancer containing four copies of ERG suggesting progression to higher ploidy. A similar scenario was observed in prostate 15481 (Figure 4).
Within individual prostates the precise pattern of ERG alteration was often complex: different patterns of ERG-rearrangement occurred juxtaposed or in close proximity (for example prostates 2007, 1266 and 543; Figure 4), or in separate regions of the same prostate slice (1266 and 543; Figure 4). The latter observation suggests that ETS gene alterations can arise independently in separate regions of the same prostate. This observation was confirmed by our discovery in an independent FISH screen of formalin-fixed trans-urethral resection of the prostate specimens of a single prostate that contained rearrangements of ETV1 and of ERG in separate foci of cancer (Figure 5).
From our analyses of a single two-dimensional slice from an entire prostate it is not possible to comment on the connection between or the independence of separate foci of cancer observed in different regions of the cancer. However in some cases the close juxtaposition of regions of cancer with distinct patterns of ERG alterations suggested possible mechanisms of cancer progression. For example, in prostate 2007 there is a region of Esplit tumour (Figure 2, lower part of left panel) that abuts a larger region of 1Edel tumour, within which resides a small region of 2+Edel tumour (see also Supplementary Figure 1). These patterns are consistent with a mechanism of progression from an initial Esplit ERG rearrangement to deletion of 5′-upstream ERG (1Edel) sequences and subsequent duplication of the TMPRSS2–ERG fusion gene to form 2+Edel. Juxtaposition of ERG rearrangements Esplit, Edel was also present in prostate 1266 (Figure 4; Supplementary Figure 4) and juxtaposition of Edel and 2+Edel were again observed in prostate 543 (Figure 4; Supplementary Figure 5). Tumour progression to a 2+Edel pattern of ERG rearrangement is significant, as we have recently shown that the presence of 2+Edel is associated with the development of very aggressive disease (Attard et al., 2007).
ERG rearrangements in pre-cancerous lesions
ERG-rearranged HG-PIN was found in four cancers (2007, 908, 1266 and 543; Figure 4; Table 1). Prostate 1266 contained five regions of HG-PIN, one with class N, two with 1Esplit and two with 1Edel patterns. Prostate 543 contained HG-PIN with 1Edel and 2+Edel patterns. We have identified a second group of pre-cancerous foci that contain ERG rearrangements. These foci have atypical epithelial appearances that fall short of a diagnosis as HG-PIN, and are consistent with a description of low grade prostatic intraepithelial neoplasia (LG-PIN). Diagnostically, these comprised clusters of non-invasive epithelial lesions in which there can be an appreciable decline of basal cells. Other features that characterize these lesions include the presence of large cuboidal luminal-type epithelial cells with granular cytoplasm and containing enlarged, pale, round-to-ovoid nuclei in which chromatin is clumped around the margin of the nuclear membrane together with nuclear stratification. Nucleoli are absent, a salient feature distinguishing these cells from HG-PIN. In comparison, HG-PIN has prominent nucleoli (>10% of nuclei) in addition to general nuclear atypia with anisokaryosis, hyperchromasia and nuclear enlargement and crowding (Epstein et al., 1995; McNeal and Yemoto, 1996; McNeal, 1988).
Low grade prostatic intraepithelial neoplasia containing an ERG rearrangement was found in four of the six prostates that contained ERG-rearranged tumour (2007, 908, 8341 and 1266; Table 1; Figure 6), with two distinct regions of 1Edel LG-PIN identified in prostate 2007 (Figure 4). In sample 1266 (Table 1; Figures 3 and 6), the immediate juxtaposition of LG-PIN, HG-PIN and tumour (all class Esplit) suggested the possible progression of LG-PIN to HG-PIN to tumour. In contrast, the 1Edel LG-PIN in prostate 2007, which has already lost 5′-ERG sequences, cannot be a precursor of the adjacent Esplit HG-PIN.
Our results provide a number of novel insights into the mechanisms of development of human prostate cancer. First we show that ETS gene alterations can arise independently in separate regions of the same cancerous prostate: particularly a single prostate was identified that harboured rearrangements of both the ERG and ETV1 genes. ERG rearrangements have been previously reported in HG-PIN (Cerveira et al., 2006; Perner et al., 2007). We have detected clonal ERG gene alterations both in HG-PIN and in atypical epithelial areas that fell short of the diagnosis of HG-PIN. The occurrence of ERG alterations in atypical epithelial areas is entirely consistent with our previous observation in which, in RT–PCR-based studies, we found TMPRSS2–ERG fusions in samples of epithelium taken from regions of the prostate that did not appear to contain either cancer or HG-PIN (Clark et al., 2006). The presence of an ERG rearrangement in LG-PIN suggests that formation of a TMPRSS2–ERG fusion gene can be an early event in prostate cancer development and that further changes may be necessary to generate HG-PIN and cancer. These results are consistent with models of cancer development in other tissues. For example, in the development of colon cancer the earliest identifiable change in the adenoma carcinoma sequence is the dysplastic crypt focus that already bears detectable APC gene mutations but has a morphology only slightly different from that of a normal polyp (Radtke and Clevers, 2005).
The presence of HG-PIN in prostate biopsies has been strongly linked to the presence of cancer (Brawer et al., 1991; Epstein et al., 1995). In contrast, no clinical relevance has been associated with the presence of lower grade lesions, such as LG-PIN (Brawer et al., 1991; Epstein et al., 1995; Egevad et al., 2006). This is largely due to problems in the consistent diagnosis of low grade lesions (Epstein et al., 1995; Srigley et al., 2000; Egevad et al., 2006) leading to the diagnostic reporting of these appearances as an entity becoming largely obsolete. The presence of an ERG rearrangement means that this subgroup of LG-PIN, which would not currently be highlighted as a worrying sign of pre-neoplasia, can now be defined. Detection of LG-PIN containing the rearranged ERG gene in a single 4 μm slice from four of six prostates with ERG-rearranged tumour suggests that the incidence of LG-PIN with an ERG rearrangement is high.
An interesting observation to emerge from these studies is that all three large transition-zone tumours examined in our study (15481, 5170 and 6972) harboured un-rearranged ERG genes. Since we were unable to find additional specimens that contained large transition-zone cancers it was not possible to test whether lack of rearranged ERG is a consistent feature of cancers contained within the transition zone. Previous comparisons of peripheral- and transition-zone cancers have given conflicting results. Cancers arising in these two zones have distinct morphological and immunohistochemical characteristics and it has been reported that transition-zone cancers are associated with more favourable pathological features (Grignon and Sakr, 1994; Greene et al., 1995; Sakai et al., 2005). In contrast, as judged by biochemical recurrence, there appeared to be no difference in outcome between transition-zone and peripheral-zone cancers following prostatectomy (Sakai et al., 2005; Chun et al., 2007). Sakai et al. (2005) found similar biochemical cure rates despite significantly higher PSA levels as well as a greater tumour volume in transition-zone cancers, and concluded that transition-zone cancers may have a less aggressive phenotype.
Studies analysing ERG alterations in prostate cancer by TMA FISH analysis or by RT–PCR categorize cancers into those that contain ERG alterations and those that do not, with some FISH reports subdividing ERG alterations into those that have retained 5′-ERG sequences and those that have not (Tomlins et al., 2005, 2006; Clark et al., 2006; Hermans et al., 2006; Iljin et al., 2006; Perner et al., 2006; Wang et al., 2006; Yoshimoto et al., 2006). We show that such classifications do not fully reflect the true complexity that exists within most cancerous prostates. Three prostates originally designated ERG rearrangement positive in TMA cores were subsequently found to contain regions of cancer with un-rearranged ERG loci, while one of four cancers designated TMA ERG rearrangement negative contained a region of cancer with an ERG rearrangement. Most prostates (4 of 5) originally designated as TMA ERG rearrangement positive actually contained more than one pattern of ERG alteration. Patterns associated with poor prognosis were not always sampled by TMA. Different patterns of ERG alteration were found together in the same region of a prostate with, in some cases, their juxtaposition suggesting possible mechanisms of cancer progression. Cancer multifocality could also contribute to the observed genetic heterogeneity. The results of our analyses on the ERG locus are consistent with previous studies of cancerous human prostates. The occurrence of prostate cancer multifocality is well documented (Villers et al., 1992; Aihara et al., 1994; Miller and Cygan, 1994; Djavan et al., 1999; Chen et al., 2000; Arora et al., 2004) and the presence of both intra- and inter-focal genetic heterogeneity has already been demonstrated at other genetic loci (Konishi et al., 1995; Mirchandani et al., 1995; Qian et al., 1995; Jenkins et al., 1997; Bostwick et al., 1998; Cheng et al., 1998).
Our data have important implications for the design of studies investigating the mechanism of development of and clinical significance of prostate cancer. Cores selected for TMA construction can miss clinically significant region of cancer and the selection of specimens without taking into account genetic heterogeneity may explain the failure of expression microarray analysis to provide a classification of prostate cancer. Our data are consistent with a model for prostate cancer development in which ERG gene alterations represent an initiating event that give rise to regions of clonal epithelial atypia. Regions of atypia/LG-PIN may then progress to form HG-PIN and cancer. By comparison very little is known about the mechanism of development of human prostate cancer that lacks ERG gene alterations and this may represent an interesting subject for future investigation.
Materials and methods
Prostate cancer specimens were obtained from a systematic series of patients who had undergone a prostatectomy at the Royal Marsden NHS Foundation Trust Hospital (RMNHSFT). Paraffin wax blocks of formalin-fixed prostatic tissue specimens were obtained from pathology archives at the referring hospitals and at the RMNHSFT. TMAs were constructed from these specimens and from TURP specimens that were taken from our cohort of conservatively managed men with prostate cancer (Cuzick et al., 2006). This study was approved by the Clinical Research and Ethics Committee at the Royal Marsden Hospital and Institute of Cancer Research.
Tissue microarrays were constructed in 35 × 22 × 7 mm blocks of Lamb paraffin wax using a manual tissue microarrayer (Beecher Instruments, Sun Prairie, WI, USA). Up to four cores of 600 μm diameter were taken from each tumour. Reassignment of each core as either ‘cancer’ or ‘normal’ was carried out on the basis of histopathological examination of H&E-stained sections that flanked the TMA slice used for FISH studies.
Tissue microarray and whole-mount sections (4 μm) were cut onto SuperfrostPlus glass slides (VWR International, Poole, UK). Each slice of whole prostate was cut and divided between two slides. FISH studies were carried out as previously reported (Attard et al., 2007). A total of 94% (31 of 33) of the samples were analysable by FISH. Probes were as follows: probe 1 (3′-ERG sequences) was a mixture of biotin-labelled bacterial artificial chromosomes (BACs) RP11-95G19, RP11-720N21 and CTD-2511E13; probe 2 (5′-ERG sequences) was DIG-labelled BACs RP11-372O17, RP11-115E14 and RP11-729O4. The ETV1 break-apart assay used biotin-labelled BACs RP11-27B1, RP11-138H16 and CTD-2008I15 corresponding to sequences 3′ to ETV1- and DIG-labelled BACs RP11-621E24, CTD-2004K7 and RP11-115D14 corresponding to sequences 5′ to ETV1. TMAs and entire tissue sections were fluorescently scanned at × 20 magnification on an Ariol SL-50 (Applied Imaging, San Jose, CA, USA) with a 5 × 0.5 μm z-stack, and images were stored for analysis. Sections were re-hybridized after stripping in 70% formamide/4 × SSC at 68 °C 4 min plus 2 × 2 min ethanol washes.
Scoring of FISH results and H&E sections
Each prostate was analysed by immunohistochemistry and FISH. First H&E and p63+AMACR-stained sections were screened so that areas of cancer, HG-PIN and other lesions could be mapped. Immunohistochemical staining was performed as described by Attard et al. (2007). The morphological criteria for selection of ‘normal’ and ‘malignant’ prostatic epithelium, HG-PIN and LG-PIN conformed to previously published definitions (Foster, 2000; Foster et al., 2000, 2004). Identification and assignment of all lesions was carried out by a review panel of three histopathologists (DB, CF and GK). Second an intervening slice cut between the H&E and p63+AMACR sections was hybridized with probes 1 and 2 (Figure 1), which detected sequences immediate 5′-(green) and 3′-(red) to the ERG gene respectively. All cells within each prostate slice were monitored for alterations in the ERG gene. Usually at least 50 epithelial nuclei per case were evaluated and in each case the modal value, usually present in at least 50% of the nuclei, was taken as the score. Where lower numbers of nuclei were examined, this is indicated in the appropriate figure legend. Where lower numbers of signals were observed in a proportion of nuclei, this reflected the fact that some nuclei were sliced and had missing signals, and was not thought to reflect pattern heterogeneity.
Preparation of RNA
Frozen prostate slices were prepared and stored in RNAlater as described by Jhavar et al. (2005). A small piece of tissue (approximately 2 mm3) identified as containing cancer on H&E-stained whole-mount sections from flanking formalin-fixed slices was washed in ice-cold phosphate-buffered saline for 5 min, embedded in OCT and frozen on dry ice. This frozen block was used to make frozen sections that were stained with H&E for reconfirmation of the presence of cancer glands. Tumour areas were then cored out from the OCT block and RNA was extracted using TRIzol (Invitrogen Ltd, Paisley, UK) as per manufacturer's instructions and was purified using the Qiagen (Crawley, UK) RNeasy MinElute clean-up kit.
This work was funded by Cancer Research UK, the National Cancer Research Institute, the Grand Charity of Freemasons and the Rosetrees Trust. We thank Christine Bell for help with typing the manuscript. DMB is supported by The Orchid Appeal.
About this article
Molecular and Cellular Endocrinology (2018)